U.S. patent application number 10/750633 was filed with the patent office on 2004-10-14 for dispersed fourier transform spectrometer.
Invention is credited to Armstrong, J. Thomas, Hajian, Arsen R., Mozurkewich, David.
Application Number | 20040201850 10/750633 |
Document ID | / |
Family ID | 33134887 |
Filed Date | 2004-10-14 |
United States Patent
Application |
20040201850 |
Kind Code |
A1 |
Hajian, Arsen R. ; et
al. |
October 14, 2004 |
Dispersed fourier transform spectrometer
Abstract
A dispersing Fourier Transform interferometer (DFTS) includes a
Fourier Transform Spectrometer having an input for receiving a
source light and an output, and a dispersive element having an
input coupled to the Fourier Transform Spectrometer output and an
output for providing the resulting multiple narrowband
interferogram outputs of different wavelengths representative of
the source light input. A processor applies a sparse sampling
algorithm for determining the best fit between a set of model
interferograms and the set of data interferograms. The model
interferogram is inferred as specified at a discrete set of lags, a
difference is determined between the model interferogram and the
data interferogram, and an optimization method determines the model
interferogram best matched to the data interferogram. The DFTS
interferometer improves the sensitivity of a standard FTS by
including a dispersive element, increasing the SNR by a factor of
(R.sub.g).sup.1/2 as compared to the FTS, where R.sub.g is the
resolving power of the conventional dispersing spectrometer (i.e.
R.sub.g=.lambda./.DELTA..lambd- a.).
Inventors: |
Hajian, Arsen R.;
(Washington, DC) ; Armstrong, J. Thomas; (Silver
Spring, MD) ; Mozurkewich, David; (Seabrook,
MD) |
Correspondence
Address: |
NAVAL RESEARCH LABORATORY
ASSOCIATE COUNSEL (PATENTS)
CODE 1008.2
4555 OVERLOOK AVENUE, S.W.
WASHINGTON
DC
20375-5320
US
|
Family ID: |
33134887 |
Appl. No.: |
10/750633 |
Filed: |
December 22, 2003 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
60435730 |
Dec 23, 2002 |
|
|
|
Current U.S.
Class: |
356/451 |
Current CPC
Class: |
G01J 3/453 20130101 |
Class at
Publication: |
356/451 |
International
Class: |
G01B 009/02 |
Claims
We claim:
1. A dispersing Fourier Transform interferometer, comprising: a
Fourier Transform Spectrometer having an input for receiving a
source light and an output; and a dispersive element having an
input coupled to the Fourier Transform Spectrometer output and an
output for providing the resulting multiple narrowband
interferogram outputs of different wavelengths representative of
the source light input.
2. An interferometer as in claim 1, further comprising a metrology
system for determining optical path lengths internal to the
interferometer.
3. An interferometer as in claim 1, further comprising: a sensor
including a plurality of light intensity sensing elements each
separately responsive to said different wavelengths for producing a
set of data of interferogram intensities I.sub.d measured at a set
of discrete lags x.sub.i; and a processor for receiving and
processing the data to produce a spectral output having a best fit
with the set of data.
4. An interferometer as in claim 3, further comprising a metrology
system for determining optical path lengths internal to the
interferometer.
5. An interferometer as in claim 3, wherein the processor includes
a sparse sampling algorithm for determining the best fit between a
set of model interferograms and said set of data
interferograms.
6. An interferometer as in claim 5, wherein the sparse sampling
algorithm comprises: processing the set of data interferograms,
I.sub.d(x.sub.i), where: 15 I d ( x i ) = s min s max s J t ( s )
cos ( 2 x i s ) ,and where s is the wavenumber, equal to the
inverse of the wavelength, J.sub.t(s) is the true spectral
intensity at wavenumber s, and the subscript t indicates that
J.sub.t(s) is the truth spectrum and is an unknown, and the
wavenumbers s.sub.min(n) and s.sub.max(n) span the range of
wavenumbers detected by the n.sup.th member of said set of light
intensity sensing elements; choosing a model spectrum,
J.sub.m(s.sub.j), from which is inferred a model interferogram
specified at a discrete set of lags x.sub.i, I.sub.m(x.sub.i); and
determining a difference between said model interferogram and said
data interferogram and applying an optimization method to determine
a model interferogram best matched to the data interferogram
I.sub.d(x.sub.i).
7. An interferometer as in claim 6, wherein the optimization method
comprises: establishing a model interferogram given by: 16 I m ( x
i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j ] cos
( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j + 1 -
s j ] . and .epsilon. is the location of a central fringe in the
model interferogram. The above expression reduces to: 17 I m ( x i
) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j = [
sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j = [ (
s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s j +
1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 18 2 = 1 n n = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 19 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( I m ( x i ) J m ( s j ) ) = i , 1 - ( i , 1
s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i , j - 1 s
j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M - 1 , ( I
m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M - 1 ) for j = M
.
8. An interferometer as in claim 6, wherein the optimization method
comprises: establishing a model interferogram given by: 20 I m ( x
i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j ] cos
( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j + 1 -
s j ] . and .epsilon. is the location of a central fringe in the
model interferogram. The above expression reduces to: 21 I m ( x i
) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j = [
sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j = [ (
s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s j +
1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 22 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 23 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( 2 ) = 2 n i = 1 n [ I m ( x i - ) - I d ( x
i ) ] ( I m ( x i - ) ) = 0 and I m ( x i - ) = 1 x i - j = 1 M - 1
( A i , j J m ( s j ) + B i , j j ) , where : A i , j = - s j + 1
cos ( z i s j + 1 ) + s j cos ( z i s j ) + sin ( z i s j + 1 ) z i
- sin ( z i s j ) z i , and B i , j = s j s j + 1 cos ( z i s j + 1
) + ( 2 s j + 1 - s j ) sin ( z i s j + 1 ) z i - s j sin ( z i s j
) z i - s j + 1 2 cos ( z i s j + 1 ) + 2 cos ( z i s j + 1 ) z i 2
- 2 cos ( z i s j ) z i 2 , where
z.sub.i=2.pi.(x.sub.i-.epsilon.).
9. An interferometer as in claim 3, wherein the source light is an
astronomical emission.
10. An interferometer as in claim 3, wherein the source light is
emitted from a material upon induction of the material into an
excited state.
11. An interferometer as in claim 3, wherein the material is an
unknown compound subjected to testing to determine the presence of
possible biologically or chemically hazardous properties.
12. As interferometer as in claim 1, wherein the Fourier Transform
Spectrometer comprises: optics for receiving and collimating a
source light along a first optical path; a beamsplitter positioned
for splitting the collimated source light into a second light beam
along a second optical path differing from said first optical path;
a first reflector positioned along said first optical path for
reflecting light transmitted through said beamsplitter back toward
a beamsplitter; a second reflector positioned along said second
optical path for reflecting said second light beam back toward a
beamsplitter; and wherein the interferometer further comprises: a
sensor including a plurality of light intensity sensing elements
each separately responsive to said different wavelengths for
producing a set of data of interferogram intensities I.sub.d
measured at a set of discrete lags x.sub.i; and a processor for
receiving and processing the data to produce a spectral output
having a best fit with the set of data.
13. An interferometer as in claim 12, wherein the processor
includes a sparse sampling algorithm for determining the best fit
between a set of model interferograms and said set of data
interferograms.
14. An interferometer as in claim 13, wherein the sparse sampling
algorithm comprises: processing the set of data interferograms,
I.sub.d(x.sub.i), where: 24 I d ( x i ) = s min s max s J t ( s )
cos ( 2 x i s ) ,and where s is the wavenumber, equal to the
inverse of the wavelength, J.sub.t(s) is the true spectral
intensity at wavenumber s, and the subscript t indicates that
J.sub.t(s) is the truth spectrum and is an unknown, and the
wavenumbers s.sub.min(n) and s.sub.max(n) span the range of
wavenumbers detected by the n.sup.th member of said set of light
intensity sensing elements; creating a continuous function
J.sub.m(s) that is equal to J.sub.m(s.sub.j) at each value s.sub.j,
from which is inferred the model interferogram specified at a
discrete set of lags x.sub.i; I.sub.m(x.sub.i); and determining the
difference between said model interferogram and said data
interferogram and applying an optimization method to determine a
model interferogram best matched to the data interferogram
I.sub.d(x.sub.i).
15. An interferometer as in claim 14, wherein the optimization
method comprises: establishing a model interferogram given by: 25 I
m ( x i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j
] cos ( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j
+ 1 - s j ] . and .epsilon. is the location of a central fringe in
the model interferogram. The above expression reduces to: 26 I m (
x i ) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j
= [ sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j =
[ ( s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s
j + 1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 27 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 28 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( I m ( x i ) J m ( s j ) ) = i , 1 - ( i , 1
s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i , j - 1 s
j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M - 1 , and
( I m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M - 1 ) for j = M
.
16. An interferometer as in claim 14, wherein the optimization
method comprises: establishing a model interferogram given by: 29 I
m ( x i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j
] cos ( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j
+ 1 - s j ] . and .epsilon. is the location of a central fringe in
the model interferogram. The above expression reduces to: 30 I m (
x i ) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j
= [ sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j =
[ ( s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s
j + 1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 31 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 32 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( 2 ) = 2 n i = 1 n [ I m ( x i - ) - I d ( x
i ) ] ( I m ( x i - ) ) = 0 and ( I m ( x i ) J m ( s j ) ) = i , 1
- ( i , 1 s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i
, j - 1 s j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M
- 1 , and ( I m ( x i ) J m ( s j ) ) = i , 1 - ( i , M - 1 s M - s
M - 1 ) for j = M . and : I m ( x i - ) = 1 x i - j = 1 M - 1 ( A i
, j J m ( s j ) + B i , j j ) , where : A i , j = - s j + 1 cos ( z
i s j + 1 ) + s j cos ( z i s j ) + sin ( z i s j + 1 ) z i - sin (
z i s j ) z i , and B i , j = s j s j + 1 cos ( z i s j + 1 ) + ( 2
s j + 1 - s j ) sin ( z i s j + 1 ) z i - s j sin ( z i s j ) z i -
s j + 1 2 cos ( z i s j + 1 ) + 2 cos ( z i s j + 1 ) z i 2 - 2 cos
( z i s j ) z i 2 , where z i = 2 ( x i - ) .
17. An interferometer as in claim 12, wherein the source light is
an astronomical emission.
18. An interferometer as in claim 12, wherein the source light is
emitted from a material upon induction of the material into an
excited state.
19. An interferometer as in claim 12, wherein the material is an
unknown compound subjected to testing to determine the presence of
possible biologically or chemically hazardous properties.
20. A dispersing Fourier Transform interferometer, comprising:
optics for receiving and collimating a source light along a first
optical path; a beamsplitter positioned for splitting the
collimated source light into a second light beam along a second
optical path substantially orthogonal to said first optical path; a
first reflector positioned along said first optical path for
reflecting light transmitted through said beamsplitter back toward
said beam splitter; a second reflector positioned along said second
optical path for reflecting said second light beam back toward said
beamsplitter; a programmable drive-train coupled to at least one of
said first and second reflectors for moving said coupled reflector
along its associated optical path so as to introduce a variable
path difference x between said first and second optical paths
whereby said source light and said second light beam recombine at
said beamsplitter and are recorded on a multielement detector at a
variety of delays, comprising an interferogram; a metrology
detector for determining the path length difference between the two
reflectors; a dispersive element positioned along said second
optical path for receiving a Fourier Transform Spectrometer output
and for providing a resulting multiple narrowband interferogram
outputs of different wavelengths representative of the source light
input; a sensor including a plurality of light intensity sensing
elements each separately responsive to said different wavelengths
for producing a set of data of interferogram intensities I.sub.d
measured at a set of discrete lags x.sub.i; and a processor for
receiving and processing the data to produce a spectral output
having a best fit with the set of data.
21. An interferometer as in claim 20, wherein the processor
includes a sparse sampling algorithm for determining the best fit
between a model interferogram and the data interferogram.
22. An interferometer as in claim 21, wherein the sparse sampling
algorithm comprises: processing the set of data interferograms,
I.sub.d(x.sub.i), where: 33 I d ( x i ) = s min s max s J t ( s )
cos ( 2 x i s ) ,and where s is the wavenumber, equal to the
inverse of the wavelength, J.sub.t(s) is the true spectral
intensity at wavenumber s, and the subscript t indicates that
J.sub.t(s) is the truth spectrum and is an unknown, and the
wavenumbers s.sub.min(n) and s.sub.max(n) span the range of
wavenumbers detected by the n.sup.th member of said set of light
intensity sensing elements; choosing a model spectrum,
J.sub.m(s.sub.j), from which is inferred a model interferogram
specified at a discrete set of lags x.sub.i; I.sub.m(x.sub.i); and
determining a difference between said model interferogram and said
data interferogram and applying an optimization method to determine
a model interferogram best matched to the data interferogram
I.sub.d(x.sub.i).
23. An interferometer as in claim 22, wherein the optimization
method comprises: establishing a model interferogram given by: 34 I
m ( x i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j
] cos ( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j
+ 1 - s j ] . and .epsilon. is the location of a central fringe in
the model interferogram. The above expression reduces to: 35 I m (
x i ) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j
= [ sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j =
[ ( s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s
j + 1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 36 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 37 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( I m ( x i ) J m ( s j ) ) = i , 1 - ( i , 1
s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i , j - 1 s
j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M - 1 , and
( I m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M - 1 ) for j = M
.
24. An interferometer as in claim 22, wherein the optimization
method comprises: establishing a model interferogram given by: 38 I
m ( x i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j
] cos ( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j
+ 1 - s j ] . and .epsilon. is the location of a central fringe in
the model interferogram. The above expression reduces to: 39 I m (
x i ) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j
= [ sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j =
[ ( s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s
j + 1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 40 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 41 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( 2 ) = 2 n i = 1 n [ I m ( x i - ) - I d ( x
i ) ] ( I m ( x i - ) ) = 0 and ( I m ( x i ) J m ( s j ) ) = i , 1
- ( i , 1 s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i
, j - 1 s j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M
- 1 , and ( I m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M - 1 )
for j = M . and : I m ( x i - ) = 1 x i - j = 1 M - 1 ( A i , j J m
( s j ) + B i , j j ) , where : A i , j = - s j + 1 cos ( z i s j +
1 ) + s j cos ( z i s j ) + sin ( z i s j + 1 ) z i - sin ( z i s j
) z i , and B i , j = s j s j + 1 cos ( z i s j + 1 ) + ( 2 s j + 1
- s j ) sin ( z i s j + 1 ) z i - s j sin ( z i s j ) z i - s j + 1
2 cos ( z i s j + 1 ) + 2 cos ( z i s j + 1 ) z i 2 - 2 cos ( z i s
j ) z i 2 , where z.sub.i=2.pi.(x.sub.i-.- epsilon.).
25. An interferometer as in claim 22, wherein the source light is
an astronomical emission.
26. An interferometer as in claim 22, wherein the source light is
emitted from a material upon induction of the material into an
excited state.
27. An interferometer as in claim 22, wherein the material is an
unknown compound subjected to testing to determine the presence of
possible biologically or chemically hazardous properties.
28. A method of determining a spectrum of a light source,
comprising: receiving and collimating a source light along a first
optical path; transmitting a first part of the collimated source
light further along said first optical path while reflecting a
second part of the collimated source light along a second optical
path; reflecting back said first part of said collimated source
light along said first optical path; reflecting back said second
part of said collimated source light along said second optical
path; introducing a path length difference x between said first and
second optical paths; recombining said back-reflected first and
second parts of said collimated source light; dispersing said
recombined light into a plurality of different wavelengths;
separately sensing an intensity I of each of said plurality of
different wavelengths to thereby produce a set of data of
interferogram intensities I.sub.d measured at a set of discrete
lags x.sub.i; and processing the data so as to produce a spectral
output having a best fit with the set of data.
29. A method as in claim 28, wherein the data processing includes
applying a sparse sampling algorithm for determining the best fit
between a model interferogram and the data interferogram.
30. A method as in claim 29, wherein the sparse sampling algorithm
comprises: processing the set of data interferograms,
I.sub.d(x.sub.i), where: 42 I d ( x i ) = s min s max s J t ( s )
cos ( 2 x i s ) ,and where s is the wavenumber, equal to the
inverse of the wavelength, J.sub.t(s) is the true spectral
intensity at wavenumber s, and the subscript t indicates that
J.sub.t(s) is the truth spectrum and is an unknown, and the
wavenumbers s.sub.min(n) and s.sub.max(n) span the range of
wavenumbers detected by the n.sup.th member of said set of light
intensity sensing elements; choosing a model spectrum,
J.sub.m(s.sub.j), from which is inferred a model interferogram
specified at a discrete set of lags x.sub.i; I.sub.m(x.sub.i); and
determining a difference between said model interferogram and said
data interferogram and applying an optimization method to determine
a model interferogram best matched to the data interferogram
I.sub.d(x.sub.i).
31. A method as in claim 30, wherein the optimization method
comprises: establishing a model interferogram given by: 43 I m ( x
i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j ] cos
( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j + 1 -
s j ] . and .epsilon. is the location of a central fringe in the
model interferogram. The above expression reduces to: 44 I m ( x i
) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j = [
sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j = [ (
s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s j +
1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 45 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 46 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( I m ( x i ) J m ( s j ) ) = i , 1 - ( i , 1
s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i , j - 1 s
j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M - 1 , and
( I m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M - 1 ) for j = M
.
32. A method as in claim 30, wherein the optimization method
comprises: establishing a model interferogram given by: 47 I m ( x
i ) = j = 1 M - 1 s j s j + 1 s [ J m ( s j ) + ( s - s j ) j ] cos
( 2 x i s ) , where : j = [ J m ( s j + 1 ) - J m ( s j ) s j + 1 -
s j ] . and .epsilon. is the location of a central fringe in the
model interferogram. The above expression reduces to: 48 I m ( x i
) = j = 1 M - 1 [ i , j J m ( s j ) + j i , j ] , where : i , j = [
sin ( 2 x i s j + 1 ) - sin ( 2 x i s j ) 2 x i ] , and i , j = [ (
s j + 1 - s j ) sin ( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s j +
1 ) - cos ( 2 x i s j ) ( 2 x i ) 2 ] setting a variance of the
residuals between the model interferogram and the data
interferogram according to the equation: 49 2 = 1 n i = 1 n [ I m (
x i ) - I d ( x i ) ] 2 . and obtaining a model interferogram best
matched to the data interferogram according to the equations: 50 2
J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( x i ) ] ( I m ( x i
) J m ( s j ) ) = 0. ( 2 ) = 2 n i = 1 n [ I m ( x i - ) - I d ( x
i ) ] ( I m ( x i - ) ) = 0 and ( I m ( x i ) J m ( s j ) ) = i , 1
- ( i , 1 s 2 - s 1 ) for j = 1 , ( I m ( x i ) J m ( s j ) ) = ( i
, j - 1 s j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j M
- 1 , and ( I m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M - 1 )
for j = M . and : I m ( x i - ) = 1 x i - j = 1 M - 1 ( A i , j J m
( s j ) + B i , j j ) , where : A i , j = - s j + 1 cos ( z i s j +
1 ) + s j cos ( z i s j ) + sin ( z i s j + 1 ) z i - sin ( z i s j
) z i , and B i , j = s j s j + 1 cos ( z i s j + 1 ) + ( 2 s j + 1
- s j ) sin ( z i s j + 1 ) z i - s j sin ( z i s j ) z i - s j + 1
2 cos ( z i s j + 1 ) + 2 cos ( z i s j + 1 ) z i 2 - 2 cos ( z i s
j ) z i 2 , where z.sub.i=2.pi.(x.sub.i-.epsilon.).
33. A method as in claim 28, wherein the source light is an
astronomical emission.
34. A method as in claim 28, wherein the source light is emitted
from a material upon induction of the material into an excited
state.
35. A method as in claim 28, wherein the material is an unknown
compound subjected to testing to determine the presence of possible
biologically or chemically hazardous properties.
Description
[0001] The present application claims the benefit of the priority
filing date of provisional patent application No. 60/435,730, filed
Dec. 20, 2002, incorporated herein by reference.
FIELD OF THE INVENTION
[0002] This invention relates to a method and device for Fourier
transform spectrometry. More particularly, the invention relates to
a spectrally dispersed Fourier Transform Spectrometer.
BACKGROUND OF THE INVENTION
[0003] The technique of spectrometry is used widely to determine
spectra either occurring in nature or in laboratory settings.
Recent advances have provided significant improvements in
spectrometric applications such as astronomical spectrometry. Prior
to 1980, measurements of Doppler velocity shifts providing velocity
precisions on the order of 1 km s.sup.-1 were seldom possible. Now,
using high precision absorption-cell spectrometers, measurement of
velocities with precisions as small as 1 m s.sup.-1 are attainted.
These data make possible the detection of planetary companions to
stars on the order of 0.16 M.sub.J<Msin(i)<15 M.sub.J
(M.sub.J is the mass of Jupiter, and i is the inclination angle of
the orbit of the planetary companion). It would be desirable,
however, to obtain a greater sensitivity capability to detect
smaller companions or to provide greater sensitivity in other types
of non-astronomical applications.
[0004] Another available technique is Fourier Transform
Spectrometry (FTS). An FTS spectrometer is an autocorrelation, or
time-domain, interferometer. The theoretical basis was laid at the
end of the 19th century (Michelson, A. A. 1891, Phil. Mag., 31,
256, Michelson, A. A. 1892, Phil. Mag., 34, 280.), but FTSs did not
achieve widespread use until approximately 75 years later (Brault,
J. W. 1985, in High Resolution in Astronomy, 15th Advanced Course
of the Swiss Society of Astrophysics and Astronomy, eds. A. Benz,
M. Huber and M. Mayor, [Geneva Observatory: Sauverny], p. 3.).
[0005] Fellgett described the first numerically transformed
two-beam interferogram and applied the multiplex method to stellar
spectroscopy (Fellgett, P., J. de Physique et le Radium V. 19, 187,
236, 1958). Fellgett employed a Michelson-type interferometer 10 as
shown in FIG. 1, wherein the incoming beam of light B is divided
into two beams B.sub.1 and B.sub.2.by a beamsplitter ("beam
divider") 12. B.sub.1 is reflected from retro reflector 14, while
B.sub.2 is reflected from retroreflector 16. As shown, beams
B.sub.1 and B.sub.2 follow separate paths whose lengths can be
precisely adjusted by delay lines (DLs) established by
repositioning one or both of reflectors 14 and 16, as shown with
reflector 16 connected to drive train 18 and drive motor 20. The
beams, now with a path difference x (i.e. the "lags"), are
recombined at beamsplitter 12 and focused by a concave mirror 22
onto a detector 24, producing an interferogram I(x), where 1 I d (
x i ) = s min s max s J t ( s ) cos ( 2 x i s ) , ( 1 ) ,
[0006] as is discussed in more detail below. A mirror 28 is
provided for an optional reference beam indicated in FIG. 1 by the
dotted lines. The reference beam is divided into two beams at the
beamsplitter 12. These two beams are reflected from the
retroreflectors 14 and 16, are recombined at 12 and focused by 22
onto the detector 24. The reference beam allows the user to align
the optics as well as to determine the zero-path position of the
DLs, i.e., the position of retroreflectors 14 and 16 for which the
path difference x=0. In this manner, the reference beam thereby
measures the delay, that is, the path difference x, introduced by
the delay lines. This provides a more accurate determination of the
optical path differences in the interferometer, and is typically
included in applications involving the precise determination of
spectral lineshapes and Doppler shifts. The detector is output to
an amplifier and demodulator 30, and then the interferogram
corresponding to the input spectrum is output to a recorder 32.
[0007] The intensity of the combined beam is measured for a series
of delay line positions. The wavelengths in the light beam cover a
range from .lambda..sub.min to .lambda..sub.max, i.e., centered on
.lambda..sub.0 and covering a range
.DELTA..lambda.=.lambda..sub.max-.lam- bda..sub.min. The most
important length parameter in the FTS is the lag, x, which is equal
to the path length difference A-B. At any given wavelength
.lambda., complete constructive interference between light from the
two paths occurs when x/.lambda. is an integer, and complete
destructive interference occurs when x/.lambda. is an integer plus
1/2.
[0008] When the paths A and B are precisely equal to within a small
fraction of .lambda..sub.0 (i.e., x=0 is the only delay for which
x/.lambda.=0 at all wavelengths), the light waves at all
wavelengths in the two beams constructively interfere and the
intensity I in the recombined beam is at its maximum, I.sub.max.
This position is known as the central fringe. As the DLs are moved
and x changes, constructive interference between light waves from
the two paths weakens, particularly at the shorter wavelengths, and
I decreases. As the magnitude of x continues to increase, I reaches
a minimum at x/.lambda..sub.0=1/2 and then rises again to a new
(but weaker) maximum at x/.lambda..sub.0=1. This weakening
oscillation of I continues as x increases. When x/.lambda..sub.0 is
increased to many times .lambda..sub.0/.DELTA..lambda- ., some
wavelengths interfere constructively and some destructively, so I
is close to the mean light level. Thus, if the observed spectral
region is wide, there is only a small range of lag with large
deviations from the mean level.
[0009] The resulting data set of intensity measurements, I(x),
measured at many values of x is known as an interferogram (Equation
1) as discussed above. The region of x over which there are large
deviations from the mean level is termed the fringe packet. An
example of a typical interferogram is shown in FIG. 2. The
wavelength of the high frequency oscillations is the central
wavelength of the bandpass, .lambda.. As shown, the number of
fringes in the central fringe packet is approximately equal to
.lambda./.DELTA..lambda., where .DELTA..lambda. is the
bandwidth.
[0010] Typically, the interferogram is sampled in steps of
.lambda..sub.0/2, and is then Fourier transformed to produce a
spectrum. The spectrum is given as a series of values at regularly
spaced discrete values of the wavelength, .lambda.. The spectrum
that results from the Fourier transform of the interferogram
contains artifacts of the PSF, which results from the finite lag
range and the actual sampling of the interferometer. A wide range
of deconvolution methods have been developed to disentangle the
real signal from the deleterious effects of sampling, noise, etc.,
and have generally done so by implicitly modeling the spectrum as
differing from zero only at discrete values of wavelength. The
disadvantage of the deconvolution approaches is that they are
highly nonlinear processes, so their behavior and uncertainties are
hard to understand quantitatively. In addition, the disadvantage of
modeling the spectrum only at discrete points is that the
corresponding interferogram has significant sidelobes.
[0011] The resolution of the spectrum at a given wavelength,
.lambda., is determined by the maximum value of x/.lambda. and can
be understood as follows. The light waves that comprise a narrow
spectral line occupy only a small range of wavelengths, and thus
stay correlated for a relatively long time, given roughly by
.delta..lambda./c, where .delta..lambda. is the range of
wavelengths making up the line and c is the speed of light. Since
the lag x corresponds to a time delay between beams of x/c, a
narrow line produces interference fringes over a large range of x.
The FTS can measure over only a finite range in x, so it cannot
distinguish between a spectral line of width .delta..lambda. and a
narrower line that produces fringes over a larger range of x. For a
spectral line wider than the resolution of the FTS, the width of a
spectral feature is measured by the range of x over which there are
interference fringes.
[0012] The most common type of spectrometer is a dispersing
spectrometer, consisting of a dispersing element (usually a
grating) and a camera equipped with an array of detectors (usually
a CCD) for multiplexing the dispersed output. Present CCD designs
allow the number of channels N.sub.ch to exceed several thousand,
so that the entire integration time is directed to integrating on
all N.sub.ch channels. Recent planetary detections have used
dispersing spectrometers with an absorption cell positioned in the
path of the incoming beam to impose a reference set of spectral
lines of known wavelength on the stellar spectrum.
[0013] In principle, an FTS offers at least three major advantages
over a dispersing spectrometer. First, the spectral resolution can
be changed simply by changing the maximum value of the lag; second,
the wavelength scale in the resulting spectrum is determined only
by the delay line settings, while remaining insensitive to such
effects as scattered light and flexure of the instrument; and
third, the point spread function (PSF) of the spectrum can be
determined to a high degree of precision.
[0014] An FTS, however, also suffers certain disadvantages. These
include low sensitivity: a conventional FTS is essentially a
single-pixel scanning interferometer, and high spectral resolution
requires measurements at a large number of lag settings.
Accordingly, FTSs are commonly used when sensitivity is not a
paramount concern, such as with laboratory spectroscopy or solar
observations, or when very high spectral resolution or accurate
wavelength calibration is required, such as in observations of
planetary atmospheres. Other applications of FTSs include FTIR,
MRI, and fluorescence and Raman emission spectroscopy.
[0015] It would therefore be desirable to provide a spectrometer
which offers the advantages of an FTS spectrometer while preserving
most of the sensitivity of a dispersing spectrometer. It would also
be desirable to provide an improved algorithm for recovering the
spectrum from the interferogram with greater fidelity, with easily
quantifiable error estimates, and without producing undesirable
artifacts.
SUMMARY OF THE INVENTION
[0016] According to the invention, a dispersing Fourier Transformn
Spectrometer (DFTS) interferometer includes a Fourier Transform
Spectrometer having an input for receiving a source light and an
output, and a dispersive element having an input coupled to the
Fourier Transform Spectrometer output and an output for providing
the resulting multiple narrowband interferogram outputs of
different wavelengths representative of the source light input.
[0017] Also according to the invention, a method of determining a
spectrum of a light source includes receiving and collimating a
source light along a first optical path; transmitting a first part
of the collimated source light further along the first optical path
while reflecting a second part of the collimated source light along
a second optical path; reflecting back the first part of the
collimated source light along the first optical path; reflecting
back the second part of the collimated source light along the
second optical path; introducing a path length difference x between
the first and second optical paths; recombining the back-reflected
first and second parts of the collimated source light; dispersing
the recombined beam into a plurality of different wavelengths;
separately sensing an intensity I of each of the plurality of
different wavelengths to thereby produce a set of data of
interferogram intensities I.sub.d measured at a set of discrete
lags x.sub.i; and processing the data so as to produce a spectral
output having a best fit with the set of data.
[0018] The data processing preferably includes applying a sparse
sampling algorithm for determining the best fit between a model
interferogram and the data interferogram. In one form, the sparse
sampling algorithm processes the set of data interferograms,
I.sub.d(x.sub.i), where: 2 I d ( x i ) = s min s max s J t ( s )
cos ( 2 x i s ) , ( 2 )
[0019] s is the wavenumber, equal to the inverse of the wavelength,
J.sub.t(s) is the true spectral intensity at wavenumber s, and the
subscript t indicates that J.sub.t(s) is the truth spectrum and is
an unknown, and the wavenumbers S.sub.min(n) and S.sub.max(n) span
the range of wavenumbers detected by the n.sup.th member of said
set of light intensity sensing elements. A model spectrum,
J.sub.m(s.sub.j), is selected, from which is inferred a model
interferogram specified at a discrete set of lags x.sub.i,
I.sub.m(x.sub.i); and a difference between the model interferogram
and the data interferogram is determined, and an optimization
method applied to determine a model interferogram best matched to
the data interferogram I.sub.d(x.sub.i).
[0020] The DFTS interferometer improves the sensitivity of a
standard FTS by including a dispersive element, increasing the SNR
by a factor of (R.sub.g).sup.1/2 as compared to the FTS, where
R.sub.g is the resolving power of the conventional dispersing
spectrometer (i.e. R.sub.g =.lambda./.DELTA..lambda.). The DFTS
interferometer disperses the recombined light beam from the FTS
module and focuses it onto a CCD detector, essentially splitting a
single broadband FTS into N.sub.ch parallel, narrowband channels. A
narrowband beam yields a spectrum with a higher signal-to-noise
(SNR) ratio than a broadband beam because a narrowband beam filters
out noise from wavelengths outside the bandpass without losing
signal from inside the bandpass. The noise in the spectrum is a
constant with a value proportional to the square root of the mean
flux level in the entire interferogram, and restricting the
bandpass and recording a narrowband interferogram filters noise
from the spectrum without losing signal.
[0021] The DFTS interferometer obtains spectra over a wide
bandpass, with an easily configurable spectral resolution that can
be very high (.lambda./.DELTA..lambda.>10.sup.6), with high
sensitivity (i.e., high spectral SNR), with a well-known PSF, and
with high velocity/wavelength precision
(.delta..lambda./.lambda.=.delta.v/c.apprxeq.10.sup.-9 using a
standard frequency stabilized HeNe laser for metrology). The
algorithm is a preferred embodiment that models the spectrum as a
continuous function rather than as a series of infinitely narrow
delta functions as is done in conventional spectral deconvolution,
and it solves the forward problem, i.e., it selects the set of
spectral intensities J.sub.m(s.sub.j) that yields an interferogram
I.sub.m(x.sub.i) that best matches the measured interferogram
I.sub.d(x.sub.i).
[0022] The combination of the DFTS interferometer and the SSA
provides more precise spectral measurements, a more precise
position of the central fringe, and greater information about noise
in the data.
[0023] Astronomical spectroscopy is one application of the DFTS
interferometer. The detection of planets requires the acquisition
of sensitive, high-resolution, high-stability spectra from their
parent stars. The minute oscillations in the Doppler velocity of
the star due to an orbiting planet have presented the most abundant
signature of extrasolar planets to date. Jupiter-mass planets
typically cause the lines in solar-type stars to be Doppler shifted
by .DELTA..lambda..sub.shift/.lambda..apprxeq.10.sup.-8. The
detection of earth-mass planets is made possible with the DFTS
interferometer when precisions improve to better than 3 m/s.
[0024] The scope of other applications is broad. Spectroscopy is
most commonly used to determine the composition of a sample. The
desired spectral features are often faint and appear over a broad
spectral range. The DFTS interferometer is ideal for such
applications since the desired spectral regions can be isolated for
maximum sensitivity. Precise knowledge of the PSF afforded by the
DFTS interferometer can permit the identification of weak features
juxtaposed to or superposed on intense background features, a
capability owing to the FTS instrumentation component.
[0025] Additional applications of the invention include but are not
limited to manufacturing and product quality control, field
detectors for hazardous compounds, Raman spectroscopy, radar
detection algorithms, and atmospheric and metallurgic
spectroscopy.
[0026] Additional features and advantages of the present invention
will be set forth in, or be apparent from, the detailed description
of preferred embodiments which follows.
BRIEF DESCRIPTION OF THE DRAWINGS
[0027] FIG. 1 is a schematic diagram of a prior art
interferometer.
[0028] FIG. 2 is a graph showing a representative conventional
interferogram.
[0029] FIG. 3 is a schematic diagram of a DFTS interferometer
according to the invention.
[0030] FIG. 4 is a spectrum showing a representative FFT applied to
a sparsely sampled interferogram.
[0031] FIG. 5 is a spectrum showing the results of applying the SSA
to the same dataset as in FIG. 4 according to the invention.
DETAILED DESCRIPTION OF THE INVENTION
[0032] Definition: The term "optics" as used herein when referring
to a component of an interferometer of the invention includes a
lens or a mirror.
[0033] Referring now to FIG. 3, a DFTS interferometer 100 includes
a beamsplitter 102 that partially reflects and partially transmits
an input light beam B, splitting it into reflected beam B.sub.1
along a first optical path and transmitted beam B.sub.2 along a
second optical path. B.sub.1 is reflected from a first
retroreflector 104 back to the beamsplitter 102, while B.sub.2 is
reflected from a second retroreflector 106 back to beamsplitter
102. As discussed above with respect to FIG. 1, the paths of either
or both of B.sub.1 and B.sub.2 can be adjusted, such as is shown
with respect to B.sub.2 where a programmable drive-train such as
that illustrated in FIG. 1 is coupled to reflector 106 and thus
introduce a path difference x between the first and second optical
paths. Beams B.sub.1 and B.sub.2 recombine at beamsplitter 102 and
due to path difference x produce an interferogram the properties of
which are dependent on the spectral content of the incident input
light beam B and on the optical path difference x.
[0034] Unlike a conventional FTS that would then focus the light on
a detector, the light from recombined beams B.sub.1 and B.sub.2 is
then directed onto a dispersing grating 108 to separate the beam
into a plurality of channels differing in wavelength as is further
described below. As illustrated in FIG. 3 for a single channel, the
light from each narrowband channel is focused by a lens 110 onto a
detector, CCD 112. By dispersing the recombined light beam from the
FTS module and focusing it onto a CCD detector, the DFTS
interferometer splits a single broadband FTS output into N.sub.ch
parallel, narrowband channels. A narrowband beam yields a spectrum
with a higher signal-to-noise (SNR) ratio than a broadband beam
because, unlike a conventional FTS spectrometer, the noise at any
given wavelength in the spectrum is not proportional to the square
root of the signal level. The noise in the spectrum is a constant
with a value proportional to the square root of the mean flux level
in the entire interferogram, and restricting the bandpass and
recording a narrowband interferogram filters noise from the
spectrum without losing signal. The DFTS interferometer realizes an
increase in the SNR by a factor of (R.sub.g).sup.1/2 as compared to
a conventional FTS. This can be shown as follows. Consider a
telescope collecting a stellar flux of W photons s.sup.-1 nm.sup.-1
(we have expressed W using units of wavenumber instead of
wavelength). An interferogram with measurements at N.sub.lag delays
is obtained with a mean level of W t.sub.lag .DELTA.s photons per
lag for a given spectral channel, where .DELTA.s=s/R.sub.g is the
bandwidth of the channel, and t.sub.lag is the integration time at
each delay. In the following analysis, we consider the data from a
single spectral channel. Since the integral of the spectral
intensities over the total spectral bandwidth is equal to the
intensity, I.sub.o, at the peak of the central fringe of the
interferogram the mean spectral intensity (i.e., the mean signal
level of the spectrum) is just I.sub.o divided by the spectral
bandwidth. Assuming that the fringe contrast is 100%, then I.sub.o
is just equal to the mean level of the interferogram, and the mean
spectral intensity is:
S.sub.s=Wt.sub.lag, (3)
[0035] On average, the noise level in the interferogram is
determined according to Poisson statistics:
.sigma..sub.I={square root}{square root over (Wt.sub.lag.DELTA.s)}.
(4)
[0036] Parceval's Theorem states that the total noise power in the
spectral and lag domains is equal: 3 S = 1 x s ( 5 )
[0037] where .sigma..sub.S is the average spectral noise power per
pixel, and .sigma..sub.I is the average noise power in the
interferogram per pixel. We combine the above equations to compute
the signal-to noise ratio in the spectrum: 4 SNR S = Wt lag s R FTS
. ( 6 )
[0038] Not surprisingly, the number of samples in the interferogram
(N.sub.lag), is directly proportional to the number of independent
spectral values, M, across one channel: 5 M = R FTS R g = 2 N lag ,
( 7 )
[0039] and Equation 6 becomes: 6 SNR S = 2 Wt lag N lag s R FTS M .
( 8 )
[0040] For the case of the conventional FTS, R.sub.g=1.
[0041] Since the width of the central fringe packet is inversely
proportional to M, small values of M mean that meaningful signal is
collected throughout a larger portion of the interferogram. A large
value of M suggests a narrow fringe will be the only region in the
interferogram that has significant signal. In effect, M serves to
dilute the signal as the fringes decorrelate. Equation 8
demonstrates that SNR.sub.S is directly proportional to
(R.sub.g).sub.1/2 for a constant integration time (t.sub.lag
N.sub.lag), source brightness (W), observing wavenumber (s), and
spectral resolving power (R.sub.FTS). Sensitivity is gained with
greater multiplexing.
[0042] FIG. 3 illustrates a DFTS interferometer 200 according to
the invention, that includes an optional metrology detector for
determining the path difference x. Two light beams enter the DFTS
interferometer, one from the source to be measured ("science
light") and the other from the laser metrology system ("metrology
light"). Laser light from the metrology laser is split into two
beams with orthogonal polarizations at BSC1. The two beams are
frequency shifted (AOMs), recombined (BSC2), and spatially filtered
and expanded (SPF) to the same size as the science light beam. Part
of the recombined beam is split from the main beam (B), both
orthogonal polarizations are mixed at the polarizer (P1) and
focused onto a reference detector (D1). Light from the source to be
measured enters the spectrometer through a polarizer (P3).
Polarized science light is combined with metrology light at the
notch filter (N1). The combined beam is split into two by a
polarizing beamsplitter (BSC3). Each beam propagates through DL1 or
DL2. The beams are recombined at a polarizing beamsplitter (BSC4).
The metrology light is separated from the combined light using a
notch filter (N2), the orthogonal polarizations are mixed with a
polarizer (P2) and sent to the metrology detector (D2). The
intensity measured at D2 is compared with that measured at D1 to
generate the metrology signal. At this point, a conventional FTS
would focus the light transmitted through N2 to a detector.
Instead, with the DFTS interferometer, the light is sent to a
dispersing spectrometer as shown. The two polarizations are
separated with a Wollaston prism (W), dispersed with a transmission
grating (G), and are each focused onto a row of pixels on the CCD
112.The data in the form of interferogram intensities, I.sub.d,
measured at a set of discrete lags, x.sub.i, where
1.ltoreq.i.ltoreq.N, is recorded on the CCD 112 with a
computer.
[0043] The DFTS interferometer in a preferred embodiment utilizes
the concept that the interferogram obtained may be "undersampled"
as compared to the sampling required using a conventional FTS.
According to the Nyquist Theorem, a FTS must be sampled at
increments .delta.x of the path difference A-B such that
.delta.x=1/(2.DELTA.s), where .DELTA.s=s.sub.max-s.sub.min is the
width in wavenumbers of the spectral region being observed. (For a
given wavelength .lambda., the wavenumber s is 1/.lambda..)
[0044] As an example, with a standard FTS the wavenumber range
.DELTA.s.sub.std is large, so the sampling interval
.delta.x.sub.std is small. The DFTS interferometer, however,
functions as R.sub.g standard FTSs working in parallel (where
R.sub.g is the number of spectral channels in the dispersing
spectrometer), each working in a narrow band. The sampling interval
.delta.x.sub.g for these narrow-band FTSs is 1/(2.DELTA.s.sub.g),
where .DELTA.s.sub.g is the wavenumber range of a single channel.
Because the channels are 1/R.sub.g as wide as the bandpass of the
standard FTS, .DELTA.s.sub.g=.DELTA.s.sub.std/R.sub.g, and the
sampling interval .delta.x.sub.g is R.sub.g times larger than
.delta.x.sub.std. Therefore, the number of sampled points needed to
attain a given resolution is reduced by a factor of up to R.sub.g
for each channel. Also, conventional FTS data processing techniques
involve converting an interferogram into a spectrum using a Fourier
transform, after which corrections are applied for sampling to
result in a final spectrum. The DFTS technique instead applies an
algorithm for inferring the best spectrum given a bandwidth limited
interferogram.
[0045] The algorithm is preferably applied as either (1) a fast
algorithm for solving the best spectrum assuming that the location
of the central fringe for each channel is known, or (2) a slow
algorithm, for solving the best spectrum as well as the best
location for the central fringe. In both approaches, the data
consist of interferogram intensities, I.sub.d, measured at a set of
discrete lags, x.sub.i, where 1.ltoreq.i.ltoreq.N. Apart from a
constant, which can be ignored in this analysis, the interferogram
is simply the inverse cosine transform of the spectrum. Therefore,
the data can be written as: 7 I d ( x i ) = s min s max s J t ( s )
cos ( 2 x i s ) , ( 9 )
[0046] where J.sub.t(s) is the spectral intensity at wavenumber s.
The subscript t indicates that J.sub.t(s) is the truth spectrum,
and is not known to the observers. It is then desired to infer
J.sub.t(s) based on observations of I.sub.d(x.sub.i).
[0047] At this point, there are two significant departures from
conventional approaches. The first is that the forward problem is
solved. The forward problem is the process of selecting the set of
spectral intensities, J.sub.m(s.sub.j), which yields an
interferogram I.sub.m(x.sub.i) that best matches I.sub.d(x.sub.i).
This is the opposite of the standard strategy of solving the
backwards problem by doing a deconvolution of I.sub.d(x.sub.i) in
the hopes of disentangling the real signal from the deleterious
effect of sampling, noise, etc., and recovering
J.sub.t(s.sub.j).
[0048] The second departure from conventional methodologies is that
a model spectrum having continuous frequency coverage is selected.
Conventional methods apply Fourier Transforms to discretely sampled
data and return discrete data. The results from conventional
methods are diminished in quality due to the lack of knowledge
between sampled frequencies. The method presented here alleviates
this problem to first order.
[0049] Initially, one starts by guessing a set of M spectral
intensities, J.sub.m(s.sub.j), which span a wavenumber range
defined by the edge wavenumbers of a single, narrowband spectral
channel. It is known that the light outside this wavenumber range
has been excluded from the detector by the conventional
spectrometer in the FTS optical train. Furthermore, it can be
assumed that the continuous spectral intensities between s.sub.j
and s.sub.j+1 are given by the interpolation between
J.sub.m(s.sub.j) and J.sub.m(s.sub.j+1)
[0050] This choice of J.sub.m(s.sub.j) and the assumptions above
result in an interferogram given by: 8 I m ( x i ) = j = 1 M - 1 s
j s j + 1 s [ J m ( s j ) + ( s - s j ) j ] cos ( 2 x i s ) , where
: ( 10 ) j = [ J m ( s j + 1 ) - J m ( s j ) s j + 1 - s j ] . ( 11
)
[0051] and .epsilon. is the location of the central fringe in the
interferogram. The integral can be evaluated analytically, reducing
the expression to: 9 I m ( x i ) = j = 1 M - 1 [ i , j J m ( s j )
+ j i , j ] , where : ( 12 ) i , j = [ sin ( 2 x i s j + 1 ) - sin
( 2 x i s j ) 2 x i ] , and ( 13 ) i , j = [ ( s j + 1 - s j ) sin
( 2 x i s j + 1 ) 2 x i ] + [ cos ( 2 x i s j + 1 ) - cos ( 2 x i s
j ) ( 2 x i ) 2 ] ( 14 )
[0052] In the fast algorithm technique, the variance of the
residuals between the model interferogram and the data
interferogram is given by: 10 2 = 1 n i = 1 n [ I m ( x i ) - I d (
x i ) ] 2 . ( 15 )
[0053] Above, when describing the forward problem, it is desired to
obtain a model interferogram best matched to the data
interferogram. This condition can be expressed as a set of
equations: 11 2 J m ( s j ) = 2 n i = 1 n [ I m ( x i ) - I d ( i )
] ( I m ( x i ) J m ( s j ) ) = 0. ( 16 )
[0054] To complete the problem requires the Jacobian, which can be
derived analytically. 12 ( I m ( x i ) J m ( s j ) ) = i , 1 - ( i
, 1 s 2 - s 1 ) for j = 1 , ( 17 ) ( I m ( x i ) J m ( s j ) ) = (
i , j - 1 s j - s j - 1 ) + i , j - ( i , j s j + 1 - s j ) for 2 j
M - 1 , ( 18 ) ( I m ( x i ) J m ( s j ) ) = ( i , M - 1 s M - s M
- 1 ) for j = M . ( 19 )
[0055] The slow algorithm technique also starts with the expression
for Equation 15. Equation 16 is still valid, and in addition: 13 (
2 ) = 2 n i = 1 n [ I m ( x i - ) - I d ( x i ) ] ( I m ( x i - ) )
= 0 ( 20 )
[0056] The Jacobian is given by Equations 17-19 and: 14 I m ( x i -
) = 1 x i - j = 1 M - 1 ( A i , j J m ( s j ) + B i , j j ) , where
: ( 21 ) A i , j = - s j + 1 cos ( z i s j + 1 ) + s j cos ( z i s
j ) + sin ( z i s j + 1 ) z i - sin ( z i s j ) z i , and : ( 22 )
B i , j = s j s j + 1 cos ( z i s j + 1 ) + ( 2 s j + 1 - s j ) sin
( z i s j + 1 ) z i - s j sin ( z i s j ) z i - s j + 1 2 cos ( z i
s j + 1 ) + 2 cos ( z i s j + 1 ) z i 2 - 2 cos ( z i s j ) z i 2 ,
( 23 )
[0057] We have used the definition
z.sub.i=2.pi.(x.sub.i-.epsilon.).
[0058] Accordingly, with the spectral reconstruction algorithm
described above, a model spectrum is first selected, resulting in a
model interferogram. The model spectrum is then varied to yield a
model interferogram that most closely matches the data
interferogram in a least-squares sense, a method that may be
implemented using a simple Newton technique.
[0059] The algorithm functions primarily as an anti-aliasing
filter, replacing the traditional Fast Fourier transform (FFT) for
our application. The spectrum resulting from a sparsely sampled
interferogram consists of the "true" spectrum plus an "aliased"
version of the spectrum shifted to other wavelengths. So long as
the sampling interval in the interferogram satisfies the Nyquist
Theorem, the aliases will not overlap the "true" spectrum. The
algorithm infers the spectral intensities only over a user-defined
bandwidth of interest (presumably containing the "real" signal and
not one of the aliases). These effects are illustrated in FIG. 4,
which shows the results of applying a FFT to a sparsely sampled
interferogram. The real spectrum (denoted by the arrow) is aliased
at all frequencies. As the sampling in the interferogram approaches
the Nyquist Limit, that is becomes sparser, the aliased peaks
merge. FIG. 5 shows the results of applying the algorithm according
to the invention for the same dataset as used with the FFT in
generating FIG. 5. It is evident from comparing the figures that
the sparse reconstruction algorithm serves as an anti-aliasing
filter, and more efficiently reconstructs the actual signal than
does the FFT. The spectrum outside this interval is zero.
[0060] The DFTS interferometer of the invention is an achromatic
device in that it can obtain spectra from a luminous source at any
given wavelengths throughout the electromagnetic spectrum, so long
as the optical components and detectors are selected so as to
provide reasonable sensitivity. Obviously many other modifications
and variations of the present invention are possible in the light
of the above teachings. It is therefore to be understood that the
scope of the invention should be determined by referring to the
following appended claims.
* * * * *